Word spotting in handwritten document images based on multiple features

Abstract

This paper presents word spotting in handwritten documents based on multiple features. Multiple features are derived using Gabor, Histogram oriented gradient (HOG), Local binary pattern, texture filters and Morphological filters. The real time documents are heterogeneous in nature, for instance application forms, postal cards, railway reservations forms etc. includes handwritten and printed text with different scripts. To spot a word in such documents and retrieving them from a huge digitized repository is a challenging task. To address such issues word spotting based on multiple features is carried out with learning and without learning methods. In both the methods (learning and learning free) texture filters are exhibiting outstanding performance in terms of precision recall and f-measures. To confirm the capability of the proposed method, extensive experiments are made on publically available dataset i.e.GW20 and noted encouraging results compared to other contemporary works.